The long-awaited EU AI Act is still under debate as European lawmakers can’t agree on how to regulate foundation models, and it’s unlikely to pass regulation before December.
As first reported by Reuters, Spain, which currently heads the EU, is pushing for more regular vetting for vulnerabilities and the creation of a tiered system of regulation depending on the number of users a model has.
European lawmakers have had three trilogues (a three-party discussion between the European Parliament, the Council of the European Union, and the European Commission) around the AI Act, with a fourth expected to happen this week.
Another meeting has been set for December, if none of the parties agree on anything this month. This has sparked concern that any decision-making on the law could be punted to next year. European lawmakers had hoped to pass the AI Act before the year ends.
One of the drafts of the EU AI Act proposes foundation model developers be required to assess potential risks, subject the models to testing throughout the development process and after market release, examine bias in training data, validate data, and publish technical documents before release.
Some open-source companies have called on the EU to consider smaller companies in the discussion. They contend that some developers may find it challenging to comply with rules, so there should be a distinction between for-profit foundation models and hobbyists and researchers.
Many government officials — including in the US — have pointed to the EU’s AI Act as a potential example to follow in drafting regulations around generative AI. But while the EU was one of the first regions to debate proposed legislation, it’s moved slower than some other international players — like China, which enacted its rules in August of this year.